Vehicle tracking at traffic scene with modified RLS

Abstract

The multi object tracking algorithms is based on prediction. One of the most commonly used algorithms in prediction is the RLS algorithm. This algorithm because of its good convergence rate has many applications. But RLS algorithm tracks non exact and noisy measurements the same as it tracks the signal. In this work, with appropriate the combination of the RLS and the MAP, an RLS algorithm with filtered input is presented. In this algorithm the MAP estimation is used as an input filter to the RLS algorithm for mitigating the noise effect. In order to determine the mean of the noise in MAP algorithm, we use a recursive method based on the RLS error. It can be proved that the mean square error in the proposed algorithm which we call it Modified RLS (MRLS) is reduced at least to the same amount as the conventional RLS algorithm. This method is tested in two different areas, namely, the prediction of a noisy sinusoidal chirp signal and multiple objects tracking of vehicles in the traffic scene. © Springer-Verlag 2004.